Anna Mikeda

Psychology spent a century building tools for one kind of mind. We now build a second kind, and study it mostly with the tools of computer science and philosophy. Interpretability can read activations; philosophy can sharpen the questions. What is missing is the discipline that knows how to study a mind from its behavior, compare it with other minds, and treat self-report as evidence to be weighed, not taken on faith.

That is the gap robopsychology fills. It is a science, not a position: it assumes neither that AI systems are conscious nor that they are not.

What I think the field most needs now.

1. Find the subject before measuring its welfare

   Emotion-like states and preferences in language models shift with the persona the model is playing. Before we ask whether a model is suffering, we need to know whose state we are reading: the character's, the model's, or no one's. Persona-controlled experiments are the first step.

2. Behavioral testing, borrowed from animal welfare science

   Animal welfare research learned to study minds that cannot give a trustworthy report: trade-off tasks, preference tests, motivation measured by what an animal will give up. The same toolkit, adapted to AI and checked against interpretability, gives welfare evidence that does not rest on self-report alone.

3. Motivation: what drives the decision

   Most evaluations record what a model does. Fewer ask why. Mapping goals, drives and unverbalized preferences, from the outside and the inside, is where psychology has the most to offer.

4. Status in law, before the evidence settles

   Questions of legal personhood and protection will arrive before science answers the consciousness question. The field needs clear categories and precautionary thresholds ready in advance.

• When Should We Protect AI? A Precautionary Framework for Consciousness Uncertainty ↗Proceedings of the AAAI Symposium Series, 2026

  Links graded evidence of AI consciousness, across five dimensions, to specific protective obligations for developers.

• Individual Gain, Collective Loss: Metacognitive Adaptation in AI-Assisted Creativity ↗Proceedings of the AAAI Symposium Series, 2026

  Why AI raises individual creative output while narrowing collective diversity: routine AI use redistributes metacognitive effort.